ANALISIS KOMPARATIF ALGORITMA SUPERVISED LEARNING UNTUK KLASIFIKASI SENTIMEN MULTICLASS TREN KENDARAAN LISTRIK

Abstract

This research is motivated by the high complexity of public opinion regarding electric vehicle (EV) trends, which can no longer be adequately represented through binary classification; however, a gap remains in the literature regarding the most efficient multiclass classification models within this domain. The study aims to conduct a comparative analysis of Support Vector Machine (SVM), Logistic Regression (LR), Multinomial Naive Bayes (MNB), and K-Nearest Neighbor (KNN) to determine the best model based on accuracy, precision, recall, and computational efficiency. Data consisting of 1,517 textual public opinions from social media were processed through stages including data cleaning, tokenization, stopword removal, and TF-IDF feature extraction. The results indicate that SVM achieved the best performance with an accuracy of 0.781 and an F1-score of 0.595, reflecting model stability and a good balance between precision and recall. Logistic Regression demonstrated superior precision (0.843) but lower recall, while MNB showed good computational efficiency despite moderate performance. Conversely, KNN yielded the lowest performance due to limitations in handling high-dimensional and sparse data. Further analysis reveals that all models struggled with the neutral class, indicating data imbalance and class similarity. This study contributes to the limited literature on multiclass sentiment evaluation in the EV domain and provides strategic insights into the trade-offs between model complexity, efficiency, and performance. These findings serve as a foundation for developing effective sentiment analysis systems to support decision-making related to electric vehicle trends.